All famous machine learning algorithms that comprise both supervised and semi-supervised learning work well only under a common assumption: the training and test data follow the same distribution. When the distribution changes, most statistical models must be reconstructed from newly collected data, which for some applications can be costly or impossible to obtain. Therefore, it has become necessary to develop approaches that reduce the need and the effort to obtain new labeled samples by exploiting data that are available in related areas, and using these further across similar fields. This has given rise to a new machine learning framework known as transfer learning: a learning setting inspired by the capability of a human being to extrapolate knowledge across tasks to learn more efficiently. Despite a large amount of different transfer learning scenarios, the main objective of this survey is to provide an overview of the state-of-the-art theoretical results in a specific, and arguably the most popular, sub-field of transfer learning, called domain adaptation. In this sub-field, the data distribution is assumed to change across the training and the test data, while the learning task remains the same. We provide a first up-to-date description of existing results related to domain adaptation problem that cover learning bounds based on different statistical learning frameworks.

A survey on domain adaptation theory: learning bounds and theoretical guarantees

Indoor/outdoor localization topic has gained a significant research interest due to the wide range of potential applications. Commonly, the Fingerprinting methods for spatial characterization of the environments monitored are employed in deterministic/statistical estimation. However, there are Fingerprint parameters that are generally neglected and can seriously affect the performance yielding to low accurate location. Nowadays, machine and deep learning (DL) methods are employed in this topic due to its ability to approximate complex non-linear models being capable of mitigating the undesirable effects of wireless propagation. In this paper, a complete overview of most influential aspects in Fingerprinting and indoor tracking methods is presented. Furthermore, a novel multi-modal complete tracking system, called SWiBluX, based on statistic and DL techniques is presented. The system relies on relevant feature extraction from available data sources to estimate user’s/target indoor position using a multi-phase statistical Fingerprint and DL disruptive approach. In addition, a Gaussian outlier filter is applied to the position estimation model output to further reduce the error in the estimation. The set of experiments performed shows that Fingerprint positioning accuracy estimation can be improved up to 45% resulting in a final estimation error that outperforms related literature.

SWiBluX: Multi-Sensor Deep Learning Fingerprint for Precise Real-Time Indoor Tracking

The unprecedented increase in the demand for mobile data, fuelled by new emerging applications and use-cases such as high-definition video streaming and heightened online activities has caused massive strain on the existing cellular networks. As a solution, the fifth generation (5G) of cellular technology has been introduced to improve network performance through various innovative features such as millimeter-wave spectrum and Heterogeneous Networks (HetNets). In essence, HetNets include several small cells underlaid within macro-cell to serve densely populated regions like stadiums, shopping malls, and so on. Recently, a mobile layer of HetNet has been under consideration by the research community and is often referred to as moving networks. Moving networks comprise of mobile cells that are primarily introduced to improve Quality of Service (QoS) for the commuting users inside public transport because QoS is deteriorated due to vehicular penetration losses and high Doppler shift. Furthermore, the users inside fast moving public transport also exert excessive load on the core network due to large group handovers. To this end, mobile cells will play a crucial role in reducing the overall handover count and will help in alleviating these problems by decoupling in-vehicle users from the core network. This decoupling is achieved by introducing separate in-vehicle access link, and out-of-vehicle back-haul links with the core network. Additionally side-haul links will connect mobile cells with their neighbors. To date, remarkable research results have been achieved by the research community in addressing challenges linked to moving networks. However, to the best of our knowledge, a discussion on moving networks and mobile cells in a holistic way is still missing in the current literature. To fill the gap, in this article, we comprehensively survey moving networks and mobile cells. We cover the technological aspects of moving cells and their applications in the futuristic applications. We also discuss the use-cases and value additions that moving networks may bring to future cellular architecture and identify the challenges associated with them. Based on the identified challenges, we discuss the future research directions.

A Comprehensive Survey on Moving Networks

This paper presents a novel software-defined wireless network architecture that integrates coordination mechanisms to enhance the capabilities of a set of central managed Wi-Fi access points (APs). The global architecture is presented in detail, where the handoff mechanism is integrated with a set of active and passive monitoring tools and other functionalities, resulting in a solution that is able to provide smart functionalities using low-cost commercial APs. The framework includes a central controller that has all the information available, and is therefore able to make smart decisions about the assignment of clients to APs. This avoids the problem of the “sticky client” that remains connected to the original AP it is associated with, rather than moving to a nearby AP, which would be a better choice. Two different test scenarios are used to compare a proactive and a reactive handoff mechanism in realistic conditions, with different walking speeds. The results illustrate the advantage of the proactive handoff, as it is more scalable and allows a better integration with other functionalities such as load balancing. The delay incurred by the handoff between APs in different channels is measured with three wireless devices, using five values for the inter-beacon time, proving that fast and seamless handoffs are possible in the scenario. The paper shows that these advanced functionalities, usually available in proprietary solutions, can also be achieved using off-the-shelf equipment.

Unsticking the Wi-Fi Client: Smarter Decisions Using a Software Defined Wireless Solution

Efficient resource allocation in enterprise wireless local area networks(WLAN) has become more paramount with the shift of traffic toward WLANs and increasing share of the video traffic. Unfortunately, current practise of client-driven association to APs has several shortcomings, e.g., sticky client problem. As a remedy, we propose to move the AP association decision to a periodically-running central controller which aims to maximize the proportionally-fair network throughput. After formulating the optimal mapping problem, we devise several heuristics requiring various degrees of knowledge, e.g., pairwise user-AP link rates, throughput demand of each user. Our analysis via simulations on realistic scenarios (conference, office, and shopping mall) shows the superior performance of our proposals in terms of aggregate logarithmic throughput. While the utility gain over the conventional client-driven approach is modest, up to 18%, the resulting increase in the weakest user's throughput is significant (71-120%) as well as that of AP load balance and fairness of user throughputs. Moreover, our evaluations reveal a very small optimality gap (between 0.1-5%). The highest gain is observed in the conference setting where the users are unevenly distributed in the network and hence there is a huge load imbalance among the APs. While schemes requiring more knowledge, i.e., on handover-cost and traffic demands, perform the best, a naive approach which runs periodically and assigns each user to the AP providing the highest signal level to that user maintains up to 41% gain in the weakest user's throughput over the client-driven handover approach.

Optimal Mapping of Stations to Access Points in Enterprise Wireless Local Area Networks

Domain Adaptation aims at benefiting from a labeled dataset drawn from a source distribution to learn a model from examples generated according to a different but related target distribution. Creating a domain-invariant representation between the two source and target domains is the most widely technique used. A simple and robust way to perform this task consists in (i) representing the two domains by subspaces described by their respective eigenvectors and (ii) seeking a mapping function which aligns them. In this paper, we propose to use Optimal Transport (OT) and its associated Wasserstein distance to perform this alignment. While the idea of using OT in domain adaptation is not new, the original contribution of this paper is twofold: (i) we derive a generalization bound on the target error involving several Wasserstein distances. This prompts us to optimize the ground metric of OT to reduce the target risk. (ii) From this theoretical analysis, we design an algorithm (MLOT) which optimizes a Mahalanobis distance leading to a transportation plan that adapts better. Experiments demonstrate the effectiveness of this original approach.

/pdf/metric-learning-in-optimal-transport-for-domain-adaptation-4yxllk176w.pdf

Metric Learning in Optimal Transport for Domain Adaptation

The Optimal transport (OT) problem and its associated Wasserstein distance have recently become a topic of great interest in the machine learning community. However, its underlying optimization problem is known to have two major restrictions: (i) it strongly depends on the choice of the cost function and (ii) its sample complexity scales exponentially with the dimension. In this paper, we propose a general formulation of a minimax OT problem that can tackle these limitations by jointly optimizing the cost matrix and the transport plan, allowing us to define a robust distance between distributions. We propose to use a cutting-set method to solve this general problem and show its links and advantages compared to other existing minimax OT approaches. Additionally, we use this method to define a notion of stability allowing us to select the ground metric robust to bounded perturbations. Finally, we provide an experimental study highlighting the efficiency of our approach.

/pdf/a-swiss-army-knife-for-minimax-optimal-transport-4rpzfaw729.pdf

A Swiss Army Knife for Minimax Optimal Transport

Even with dense base station deployments, public transport users often have a low quality for mobile services. Due to the insulation of the vehicle, passengers experience little to no connectivity on their end devices and low data bitrate. We thus propose a mobile relay architecture based on two imbricated levels of LTE networks and adapted to public transport systems. This architecture is evaluated on an actual LTE test-bed, consisting in standard radio interfaces, off-the-shelf terminals and the Amarisoft software suite. One of the specifics of a mobile relay is the use of a radio link between the core network and the base stations, with all the limitations of this medium. This work therefore evaluates the impact on data rate of signaling mechanisms and of tunneling in that context.

/pdf/mobile-relay-for-lte-proof-of-concept-and-performance-o31dvdhtjk.pdf

Mobile Relay for LTE: Proof of Concept and Performance Measurements

The current and future IEEE 802.11 deployment could potentially offer wireless Internet connectivity to mobile users. The limited AP radio coverage forces mobile devices to perform frequent handovers while current operating systems lack efficient mechanisms to manage AP transition. Thus we propose an anticipation-based handover solution that uses a Kalman filter to predict the short term evolution of the received power. This mechanism allows a mobile device to proactively start scanning and executing a handover as soon as better APs are available. We implement our mechanism in Android and we show that our solution provides a better wireless connection.

/pdf/handover-triggering-in-ieee-802-11-networks-32kikauw13.pdf

Handover triggering in IEEE 802.11 networks

Faced by the large number of deployed Wifi Access Points (AP), many research efforts focus on energy savings in Wireless Local Networks. One of the most promising solutions for improving energy efficiency is the Sleep Mode approach, which is especially effective in dense deployments. It is based on switching off the APs while they are not in use, in order to avoid unnecessary energy consumption. In this paper we evaluate the potential of switching off APs using real measurements taken in a dense urban area. We collected traces for more than 20 hours, confirming the high density of currently deployed APs in such an environment. Based on these traces, we evaluate how many APs can be switched off while maintaining the same coverage. To this end, we propose two algorithms that select the minimum set of APs needed to provide full coverage. We compute several performance parameters, and evaluate the proposed algorithms in terms of the number of selected APs, and the coverage they provide. Our results show that between 4.25% and 10.91% of the detected APs are sufficient to provide the same coverage, depending on the data set, the mobile terminal and the AP selection algorithm.

/pdf/minimal-access-point-set-in-urban-area-wifi-networks-34c5f59ccp.pdf

Tanguy Kerdoncuff

Papers

Metric Learning in Optimal Transport for Domain Adaptation

A Swiss Army Knife for Minimax Optimal Transport

Mobile Relay for LTE: Proof of Concept and Performance Measurements

Handover triggering in IEEE 802.11 networks

Minimal access point set in urban area Wifi networks